2.系统环境、编译与运行

优质

小牛编辑

131浏览

2023-12-01

Host

环境	规格
操作系统	Ubuntu 18.04
构建系统	Catkin Build System
CUDA(optional)	10.2
TensorRT(optional)	7.0.0
libtorch	cxx11版本
pytorch	1.5
onnx	1.1
netron（optional）	/

Device

环境	规格
操作系统	ubuntu18.04
jetpack工具包	JetPack 4.4
CUDA	10.2
TensorRT	7.0.0
OpenCV	4.2

编译

首先安装相应依赖。建议使用ubuntu 18.04

sudo apt-get install -y ros-melodic-opencv3             \
                        ros-melodic-cv-bridge           \
                        ros-melodic-image-transport     \
                        ros-melodic-stage-ros           \
                        ros-melodic-map-server          \
                        ros-melodic-laser-geometry      \
                        ros-melodic-interactive-markers \
                        ros-melodic-tf                  \
                        ros-melodic-pcl-*               \
                        ros-melodic-libg2o              \
                        ros-melodic-rplidar-ros         \
                        ros-melodic-rviz                \
                        protobuf-compiler               \
                        libprotobuf-dev                 \
                        libsuitesparse-dev              \
                        libgoogle-glog-dev              \

Libtorch 安装 libtorch 是 pytorch 提供的 c++ 版本库，基于caffe2 aten和c10，我们可以直接安装编译好的版本，需要注意的是libtorch 有两个版本，其一采用c11，另一个使用c11以前标准，这个版本可能不能和ros一起编译。因此需要注意一下。此处不要求帧率，所以就采用纯cpu推理

wget https://download.pytorch.org/libtorch/cpu/libtorch-cxx11-abi-shared-with-deps-1.6.0%2Bcpu.zip
unzip libtorch.zip
sudo mv libtorch /usr/local/

CUDA 10.2安装 NVIDIA CUDA10.2

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/cuda-ubuntu1804.pin
sudo mv cuda-ubuntu1804.pin /etc/apt/preferences.d/cuda-repository-pin-600
wget http://developer.download.nvidia.com/compute/cuda/10.2/Prod/local_installers/cuda-repo-ubuntu1804-10-2-local-10.2.89-440.33.01_1.0-1_amd64.deb
sudo dpkg -i cuda-repo-ubuntu1804-10-2-local-10.2.89-440.33.01_1.0-1_amd64.deb
sudo apt-key add /var/cuda-repo-10-2-local-10.2.89-440.33.01/7fa2af80.pub
sudo apt-get update
sudo apt-get -y install cuda

TensorRt 7.0.0 安装 NVIDIA TensorRT 7.0.0 直接在官方网站下载deb包，解压缩文件，完成安装。

使用说明

using SSD-detection

train

set your dataset(voc) path which contains images, annotation directory in ./utils/setting_dict.py

"test" : {
            "data_set" : ["/DJI/DJItest"],
            "batch_size" : 2,
            "transform" :
                {
                    "PIXEL_MEAN" : [123, 117, 104],
                    "IMAGE_SIZE" : 512,
                }
        },

   "train": {
            "data_set" :  ["/DJI/DJItrain/"],
            "batch_size" : 8,
            "transform" :
                {
                    "PIXEL_MEAN" : [123, 117, 104],
                    "IMAGE_SIZE" : 512,
                },

run the train.py

python3  train.py

(optional) specify the output path your can view the loss using tensorboard if you like , the loss will be stored in output directory

python3 train.py  --out_dir {YOUR_PATH}

test

python3 test.py

finetune

python3 train.py --fine_tune 1  --pretrained_model "{PRETRAINED_MODEL_PATH}"

using RTS-deploy

rosrun ICRA-vision ICRA_vision